Effective price increases are occurring one after another for AI services such as “GitHub Copilot” and “Claude.” The true cost of generative AI, previously hidden in subscription fees, has finally begun to appear on corporate invoices.

　Fixed monthly fees were like time bombs that would eventually spike once fully utilized. Now multiple companies are beginning to measure token costs on the same footing as labor costs. This is not a question of technology, but of how to allocate corporate resources between people and AI, and who will make the decisions by what criteria.

　This awareness of the problem is not limited to a handful of companies: LayerX, a back-office SaaS; Raksul, an expense-management SaaS; Sansan, which expanded its business from business-card management; freee, an accounting cloud; and Mercari, a secondhand marketplace app. From the AI and HR executives at the five companies we interviewed, we heard remarkably similar tones in their discussions of token costs.

Table of Contents
  • 1. The Subscription That Was a Time Bomb
  • 2. Tokens and Labor Costs on the Same Scale
  • 3. Not “We Won’t Use It Because It’s Expensive,” but “How Can We Use It More Cheaply?”
  • 4. The Chorus of “I Have No Intention of Cutting Jobs” and the Replacement Already Underway
  • 5. Why Top HR Executives Are Looking at AI

The Subscription That Was a Time Bomb

　It started with price increases. GitHub Copilot, a coding-assistance service, revised its pricing structure, while the U.S. company Anthropic’s Claude also made it easier to burn through tokens with its higher-tier models and new features. When companies began seriously integrating AI agents into their operations, the bills jumped sharply. Costs that had been difficult to see during the subscription period were becoming clearer as pricing moved closer to pay-as-you-go.

itmGam.refresh('InArtSpecialLink');

　In June, LayerX CEO Yoshinori Fukushima wrote on the company’s note page, “From Token Maxing to Token Management.” His point was that the focus was shifting from the stage of using AI without restraint to the stage of asking whether the tokens used had been properly converted into business value—and that, beyond the point at which using AI wholesale had become a prerequisite for competition, the next question would be ROI.

　The scale of these costs is impossible to ignore, either. Yuta Oma, Sansan’s CHRO (Chief Human Resources Officer) and CAXO (Chief AI Transformation Officer), says, “We expect AI costs to grow several-fold or more year over year.” If you are merely using chat to look things up, the cost will not reach anything like the level of labor costs.

　But running AI agents routinely is a different story. Costs quickly enter the territory of “the equivalent of one employee.” The assumption that tokens should be used “as much as possible” began to waver simultaneously across companies.